Electronic monitoring (EM) systems using onboard cameras are increasingly common in global fisheries. Artificial intelligence (AI) and machine learning (ML) tools are now being trialled to analyse EM footage, with the potential to deliver faster, more accurate identification and quantification of catch. This project explores the use of AI for catch and bycatch detection in Australia's sub-Antarctic fisheries, including those around Heard Island and McDonald Islands (HIMI), Macquarie Island, and the Ross Sea, where toothfish (Dissostichus spp) are the primary target.
We present initial results from AI models trained to detect and count both target and bycatch species, such as grenadiers (Macrouridae), deepwater skates (Bathyraja spp.), and morid cods (Antimora rostrata). The HIMI fishery, which operates using longlines and holds Marine Stewardship Council (MSC) certification, provides a rigorous testing ground for these technologies. We will present results that detect, count and estimate the length of the target species and compare results with on-board observers and logbooks. We also explore the feasibility of real-time AI-based monitoring. This technology could significantly enhance fisheries monitoring while reducing costs for both industry and regulators. Challenges in training models, handling varied visual conditions, and ensuring accuracy across species are also discussed.